Research on fluency

Fluency is the third component of the CAF triad, and similar to complexity, it has a multifaceted nature (Pallotti, 2009; Tavakoli, 2016). I discussed in the previous chapter that Lennon (1990, 2000) regarded proficiency in a ‘broad’ and in a ‘narrow’ sense. The broad definition of fluency views it as general proficiency or the ability to communicate efficiently in a language. However, the second definition may concern us more as applied linguistics researchers and/or language testers (or whatever SLA identity to which we associate ourselves). In the narrow sense, according to de Jong, Groenhout, Schoonen, and Hulstijn (2015), fluency is “described in terms of speedy and smooth delivery of speech without (filled) pauses, repetitions, and repairs” (p. 224). Some researchers who study fluency in L2 development and assessment believe that, in order to arrive at a reliable account of what constitutes fluent L2 speech, one should analyze data from L1 and L2 alongside each other (e.g. Segalowitz, 2010). de Jong and his colleagues divided the construct of fluency into three main categories of cognitive, utterance, and perceived fluency:

“[C]ognitive fluency [is] the ability of the speaker to smoothly translate thoughts to speech. However, this ability cannot be measured directly.

37

Therefore, researchers use measures of utterance fluency to gauge speech- planning difficulties that surface in utterances by counting the number of filled pauses, corrections, and repairs, and by measuring the duration of pauses. Yet another sense of fluency is perceived fluency, which pertains to the inference listeners (raters) make on the basis of the utterance about speakers’ ability (about speakers’ cognitive fluency)” (de Jong et al., 2015, p. 225).

de Jong et al. classified utterance fluency into speed fluency that accounts for the rate of the speech, breakdown fluency, that is silence and pausing, and repair fluency which deals with reformulations and hesitations. Figure 3 summarizes different types of fluency based on de Jong et al (2015) and Segalowitz (2010):

Figure 3. Typology of speech fluency (Based on de Jong et al., 2015; Segalowitz, 2010)

Fluency

Cognitive fluency

Utterance fluency

Speed fluency Breakdown

fluency

Repair fluency

Perceived fluency

38

In a rather different framework, Skehan (2014) suggested that there is a difference between the nature of fluencies measuring speakers’ speed and those focusing on the flow in their speech. In other words, disfluencies that interrupt the flow are to be distinguished from disfluencies that affect the speed (Tavakoli, Campbell, & McCormack, 2016). This distinction in fluency types is displayed in Figure 4:

Figure 4. Skehan's (2014) framework of speech fluency

Several studies, some of which are reviewed below, have tried to find the most reliable measures to assess utterance fluency (i.e. the most frequently investigated type of fluency) in L2 development and performance. According to Tavakoli (2016) and Witton-Davies (2014), it seems that the following measures can be reliably used to analyze utterance fluency:

• Pausing

o Length of pause o Frequency of pauses

o location of the pause in the clause • Speed o Speech rate Fluency Speed Speech rate Flow Pausing Reformulations

39 o Articulation rate

• Phonation time (i.e. speaking time minus pauses)

• Mean length of run (i.e. mean number of syllables between pauses)2 • Repair measures (e.g. number of hesitations, reformulations, etc.)

Not a lot of studies have focused on either some or all of CAF components within the ITA framework. In one of such studies, Gorsuch (2011) emphasized the significance of fluency in helping ITAs succeed in their performance. She operationalized fluency in terms of “intact” vs. “split” pause groups where phrasal boundaries are syntactically observed as in the former and broken as in the latter (p. 3). In her study, Gorsuch compared the influence of repeated reading (RR) as the instructional input throughout an ITA training course on the improvement of ITA candidates’ fluency against a production-oriented control group. The results indicated that the RR input group used fewer split pause groups at the end of the instruction period. Gorsuch suggested that ITAs that are more fluent could plan at the discourse level to produce more intact pause groups. Overall, the study results advocated a blend of orientation and input-focused designs for ITA training courses (p. 5).

In a study on fluency measures, de Jong et al. (2015) investigated the role of L2 fluency measures in predicting L2 proficiency. Moreover, they studied whether “L2 fluency measures that are corrected for L1 fluency behavior” can predict L2 proficiency better than uncorrected measures (p. 226). Coupled with AS-units, certain measures of utterance fluency were used including mean duration of syllables (speed fluency), number of silent pauses, length of silent pauses, and number of nonlexical filled pauses (breakdown fluency), and number of repetitions

2 Tavakoli (2016) classified phonation time and articulation rate as composite measures of fluency, that is, measures that track both speed and pausing in speech.

40

and number of corrections (repair fluency). de Jong and his colleagues found that only some of the fluency measures correlated with proficiency level. This was contrary to the findings of de Jong, Steinel, Florijn, Schoonen, and Hulstijn (2013) in which all measures of proficiency turned out to be related to L2 proficiency. In de Jong et al.’s (2015, p. 238) study, mean syllable

duration was “strongly” correlated with proficiency in L2, whereas pause duration and

proficiency did not significantly correlate. Additionally, the results revealed that “for the fluency measure syllable duration, a corrected score is more strongly related to a measure of L2

proficiency than is the original uncorrected L2 measure” (p. 237). Overall, concerning the role of L1, de Jong et al. (2015, p. 239) concluded that language tests can benefit from using “L1

behavior as a baseline” in their design, and learners can also benefit from modifying “their speaking style” either in L1 or L2.

Mirdamadi and de Jong (2015) compared the effect of syntactic complexity on utterance fluency in both L1 and L2. The researchers chose active and passive structures based on the hypothesis that, since passive voice is acquired later than active constructions, “producing a passive [would] be more difficult than producing an active… [i]t is likely that the passive

structure is not as proceduralized and automatized as the active structure” (p. 108). College-level L1 speakers of Dutch reacted to cartoon images producing 40 sentences in English and 40 sentences in Dutch. Thus, Mirdamadi and de Jong employed articulation rate (a speed fluency measure) and number of hesitations (a hybrid measure of repair and breakdown fluencies). Analysis of the recorded passive and active sentences indicated that complex language (i.e. language including more passives) affected fluency by causing more hesitations in both L1 and L2 speech. It is worth noting that fluency in L1 was negatively influenced more than L2 fluency. Nonetheless, speech rate remained unaffected by syntactic complexity.

41

Tavakoli et al. (2016) studied the pedagogical aspects of L2 fluency development and its relationship with complexity and accuracy. The experimental group in the study, who were ESL students in an EAP course, experienced “awareness-raising activities” (e.g. listening to nonnative speakers’ speech samples) and fluency strategy training” (e.g. teaching the use of gap fillers) in the course of four weeks (p. 453). Analysis of the first minute of the participants’ monologic performance indicated that although the control group gained improvement in fluency, this improvement was much more significant for the experimental group. This improvement was mainly observed in fluency measures of “length of run, articulation and speech rates, and phonation time ratio” (p. 463). Furthermore, as Tavakoli et al. reported, “the development of breakdown fluency (i.e. silence and pausing) is slower and less sensitive to pedagogic

intervention” (p. 464). Surprisingly, in connection with complexity and accuracy, it was only the control group who significantly improved in a specific measure of accuracy (i.e. percentage of error-free clauses). By contrast, the experimental group showed partial progress in these two constructs.

In a very recent article, Tavakoli (2016) scrutinized current measures of L2 fluency through comparing monologues and dialogues. She claimed the construct itself needs to be redefined, since research on fluency is overshadowed by “mixed results due to the lack of a systematic approach to measuring fluency” (p. 134). She argued that research on monologic tasks has been abundant due to the relative ease of analyzing data from such tasks, whereas the difficulty associated with measuring linguistic aspects of speech in dialogic tasks has limited their use in data collection. Following Tavakoli et al. (2016, see Chapter I), Tavakoli suggested that selecting fluency measures should be done carefully because of a possible overlap between some of these measures (see phonation time and mean length of run earlier in this section).

42

Finally, Tavakoli introduced two “dialogue only measures” in her article that can be used to describe fluency in a dialogic task (p. 139): number of turns and number of interruptions.